rfdiffusion

De novo protein backbone generation for binder design uses a diffusion-based generative model to create novel structures. You provide a target PDB, specify contigmap configurations and hotspot residues, and execute the model to generate the desired backbones.

RFdiffusion supports motif scaffolding by allowing motif insertion into generated backbones, and it handles symmetric oligomer designs for multi-chain symmetric assemblies. These features enable complex de novo protein structure creation for diverse research applications.

RFdiffusion backbone generation requires Python 3.9 or higher and CUDA 11.7 or higher support. You also need to provide contigmap and hotspot specifications to configure the diffusion model for generating target protein structures.

After generating protein backbones, you pair the output with ProteinMPNN for sequence design and use Alphafold or Chai for structure validation. ProteinQC is then used for quality control filtering to ensure the final designs meet structural criteria.

You can run RFdiffusion locally using python run_inference.py with your contigs and hotspot arguments, or execute it via cloud environments using modal run modal_rfdiffusion.py, passing the target PDB and design specifications to generate backbones.

Contigmap and hotspot specifications are required to guide the diffusion model in generating backbones that correctly interface with the target. Contigs define the structural constraints and chain lengths, while hotspots specify the exact target residues for binder interaction.